SPD: Sync-Point Drop for efficient tensor parallelism of Large Language Models

Open in new window